November 2008 - Mpich2-core - mailman3.cels.anl.gov

Proposal for Out-of-band messaging in PMI-v2
by Pavan Balaji 24 Nov '08

24 Nov '08

Hi all, I've added a proposal for out-of-band messaging in PMI-v2 (http://wiki.mcs.anl.gov/mpich2/index.php/PMI_v2_Wire_Protocol) We should probably have a separate page for proposals, but it's on the main page for the time being. The idea of OoB messaging is to allow the PMI server to asynchronously tell the MPI process to do some task. For example, this can be useful in user-directed checkpoint or graceful abort when one process call MPI_Abort. Please take a look at it and let me know your comments. Thanks. -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

make -j
by David Gingold 20 Nov '08

20 Nov '08

If I try to build MPICH2 with "make -j 6" etc, the build generally falls over in one way or another. Is this a known problem? -dg

3 4

MPICH2 BoF Slides
by Darius Buntinas 20 Nov '08

20 Nov '08

I'm attaching the first cut of the slides. I based these on last year's slides and some from Rajeev. What did I miss in the current status? Pavan, do you want to send me a slide for Hydra? If not can you give me a few bullets to put under "Hydra" in the current status slide? BTW, Alexander can't make it, but he was going to ask someone from the Blue Waters MPI team to present his slides. I haven't heard more on that. I've also asked Howard from Cray to present (he agreed), and Larry from SiCortex. Larry indicated that he's interested, but hasn't committed yet. Thanks, -d

3 9

DCMF BoF slides
by Pavan Balaji 18 Nov '08

18 Nov '08

I've attached a draft of the DCMF BoF slides. It's pretty high-level. Is there anything else I need to talk about? -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 1

Re: [mpich2-core] (no subject)
by Anthony Chan 17 Nov '08

17 Nov '08

Just built mpich2-1.1a2, all documentations seem to be there, so are the recently added tests. Ready to go from my end. A.Chan ----- "Pavan Balaji" <balaji(a)mcs.anl.gov> wrote: > I've tagged the current version of trunk and placed a tarball here: > /homes/balaji/mpich2-1.1a2.tar.gz (on MCS). > > Anthony -- can you try out this version and let us know? > > Thanks. > > -- Pavan > > On Sun, 16 Nov 2008, Anthony Chan wrote: > > > > > I don't have anything to check in for this release. > > > > A.Chan > > ----- "Rajeev Thakur" <thakur(a)mcs.anl.gov> wrote: > > > >> Should we cut a release tarball for testing or does anyone have > >> anything to > >> checkin? > >> > >> I guess we can ignore the SMPD-Nemesis thread test failures for > now. > >> We plan > >> to have an alpha3 in the next few weeks anyway, to include Nemesis > on > >> Windows and Nemesis-PSM (Qlogic). > >> > >> Rajeev > > > > -- > Pavan Balaji > http://www.mcs.anl.gov/~balaji

4 11

(no subject)
by Rajeev Thakur 16 Nov '08

16 Nov '08

Should we cut a release tarball for testing or does anyone have anything to checkin? I guess we can ignore the SMPD-Nemesis thread test failures for now. We plan to have an alpha3 in the next few weeks anyway, to include Nemesis on Windows and Nemesis-PSM (Qlogic). Rajeev

4 4

Solaris voodoo
by Pavan Balaji 14 Nov '08

14 Nov '08

All, I was trying to debug some code on Solaris and ran into some weird behavior. For a simple program which does: int main() { return 1; } $ gcc foo.c -o foo && ./foo; echo $? 1 --> gives the return value of the program The same thing on solaris: $ cc foo.c -o foo && ./foo; echo $? 0 --> success?! Any ideas what I'm missing here? -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 2

Re: [mpich2-core] Release
by Anthony Chan 14 Nov '08

14 Nov '08

Yes, I've checked it. I would consider this be very low risk (as long as you don't set MAKE_CFLAGS). As Bill said, README, installguide, and wiki will need to be updated. BTW, I am thinking of using MPICH2_MAKE_CFLAGS instead of just MAKE_CFLAGS to avoid any potential name conflict.... A.Chan ----- "Rajeev Thakur" <thakur(a)mcs.anl.gov> wrote: > If it is low risk and will not break anything. > > Rajeev > > > -----Original Message----- > > From: mpich2-core-bounces(a)mcs.anl.gov > > [mailto:[email protected]] On Behalf Of Anthony Chan > > Sent: Friday, November 14, 2008 2:24 PM > > To: mpich2-core(a)mcs.anl.gov > > Subject: Re: [mpich2-core] Release > > > > Does it mean that the "MAKE_CFLAGS" can still be put in > > the simplemake.in ? > > > > A.Chan > > > > BTW, my version of simplemake support MAKE_CFLAGS. With the > > latest svn source, when configured with > > CC=gcc CXX=g++ F77=ifort F90=ifort, the make step > > 'make VERBOSE=1 MAKE_CFLAGS="-Wall -Werror"' fails as follows: > > > > gcc -DHAVE_CONFIG_H -I. > > -I/home/chan/mpich_work/mpich2/src/mpid/common/datatype > > -I/home/chan/mpich_work/mpich2/src/mpid/common/datatype > > -I/disks/chan/mpich2_work/build_tmp/src/include > > -I/home/chan/mpich_work/mpich2/src/include -O2 -Wall -Werror > > -DUSE_PROCESS_LOCKS > > -I/disks/chan/mpich2_work/build_tmp/src/mpid/ch3/include > > -I/home/chan/mpich_work/mpich2/src/mpid/ch3/include > > -I/disks/chan/mpich2_work/build_tmp/src/mpid/common/datatype > > -I/home/chan/mpich_work/mpich2/src/mpid/common/datatype > > -I/disks/chan/mpich2_work/build_tmp/src/mpid/common/locks > > -I/home/chan/mpich_work/mpich2/src/mpid/common/locks > > -I/disks/chan/mpich2_work/build_tmp/src/mpid/ch3/channels/neme > sis/include -> > I/home/chan/mpich_work/mpich2/src/mpid/ch3/channels/nemesis/in > > clude > > -I/disks/chan/mpich2_work/build_tmp/src/mpid/ch3/channels/neme > sis/nemesis/include -> > I/home/chan/mpich_work/mpich2/src/mpid/ch3/channels/nemesis/ne > > mesis/include > > -I/disks/chan/mpich2_work/build_tmp/src/mpid/ch3/channels/neme > sis/nemesis/utils/monitor -> > I/home/chan/mpich_work/mpich2/src/mpid/ch3/channels/nemesis/ne > > mesis/utils/monitor > > -I/disks/chan/mpich2_work/build_tmp/src/mpid/common/locks > > -I/home/chan/mpich_work/mpich2/src/mpid/common/locks -c > > /home/chan/mpich_work/mpich2/src/mpid/common/datatype/mpid_typ > > e_debug.c > > cc1: warnings being treated as errors > > /home/chan/mpich_work/mpich2/src/mpid/common/datatype/mpid_typ > > e_debug.c:480: warning: 'MPIDI_Datatype_depth_spacing' > > defined but not used > > make[5]: *** [mpid_type_debug.o] Error 1 > > make[5]: Leaving directory > > `/disks/chan/mpich2_work/build_tmp/src/mpid/common/datatype' > > make[4]: *** [all-redirect] Error 2 > > make[4]: Leaving directory > > `/disks/chan/mpich2_work/build_tmp/src/mpid/common/datatype' > > make[3]: *** [all-redirect] Error 2 > > make[3]: Leaving directory > > `/disks/chan/mpich2_work/build_tmp/src/mpid/common' > > make[2]: *** [all-redirect] Error 2 > > make[2]: Leaving directory > > `/disks/chan/mpich2_work/build_tmp/src/mpid' > > make[1]: *** [all-redirect] Error 2 > > make[1]: Leaving directory `/disks/chan/mpich2_work/build_tmp/src' > > make: *** [all-redirect] Error 2 > > > > ----- "Rajeev Thakur" <thakur(a)mcs.anl.gov> wrote: > > > > > Pavan is already en route to Austin. We need to wait at least one > > > night for > > > a clean set of tests. So sometime over the weekend or Monday. > > > Technically we > > > have until Thursday to release it (our BoF is on Thursday). > > > > > > Rajeev > > > > > > > > > > > > _____ > > > > > > From: mpich2-core-bounces(a)mcs.anl.gov > > > [mailto:[email protected]] On Behalf Of Jayesh > Krishna > > > Sent: Friday, November 14, 2008 12:39 PM > > > To: balaji(a)mcs.anl.gov > > > Cc: mpich2-core(a)mcs.anl.gov > > > Subject: [mpich2-core] Release > > > > > > > > > > > > Pavan, > > > When are we cutting the tarball for the release ? > > > > > > (PS: Is Hydra good to go ? Did see some commits from you related > to > > > Hydra > > > last night...) > > > > > > Regards, > > > Jayesh > >

1 0

Re: [mpich2-core] Release
by Anthony Chan 14 Nov '08

14 Nov '08

Does it mean that the "MAKE_CFLAGS" can still be put in the simplemake.in ? A.Chan BTW, my version of simplemake support MAKE_CFLAGS. With the latest svn source, when configured with CC=gcc CXX=g++ F77=ifort F90=ifort, the make step 'make VERBOSE=1 MAKE_CFLAGS="-Wall -Werror"' fails as follows: gcc -DHAVE_CONFIG_H -I. -I/home/chan/mpich_work/mpich2/src/mpid/common/datatype -I/home/chan/mpich_work/mpich2/src/mpid/common/datatype -I/disks/chan/mpich2_work/build_tmp/src/include -I/home/chan/mpich_work/mpich2/src/include -O2 -Wall -Werror -DUSE_PROCESS_LOCKS -I/disks/chan/mpich2_work/build_tmp/src/mpid/ch3/include -I/home/chan/mpich_work/mpich2/src/mpid/ch3/include -I/disks/chan/mpich2_work/build_tmp/src/mpid/common/datatype -I/home/chan/mpich_work/mpich2/src/mpid/common/datatype -I/disks/chan/mpich2_work/build_tmp/src/mpid/common/locks -I/home/chan/mpich_work/mpich2/src/mpid/common/locks -I/disks/chan/mpich2_work/build_tmp/src/mpid/ch3/channels/nemesis/include -I/home/chan/mpich_work/mpich2/src/mpid/ch3/channels/nemesis/include -I/disks/chan/mpich2_work/build_tmp/src/mpid/ch3/channels/nemesis/nemesis/include -I/home/chan/mpich_work/mpich2/src/mpid/ch3/channels/nemesis/nemesis/include -I/disks/chan/mpich2_work/build_tmp/src/mpid/ch3/channels/nemesis/nemesis/utils/monitor -I/home/chan/mpich_work/mpich2/src/mpid/ch3/channels/nemesis/nemesis/utils/monitor -I/disks/chan/mpich2_work/build_tmp/src/mpid/common/locks -I/home/chan/mpich_work/mpich2/src/mpid/common/locks -c /home/chan/mpich_work/mpich2/src/mpid/common/datatype/mpid_type_debug.c cc1: warnings being treated as errors /home/chan/mpich_work/mpich2/src/mpid/common/datatype/mpid_type_debug.c:480: warning: ‘MPIDI_Datatype_depth_spacing’ defined but not used make[5]: *** [mpid_type_debug.o] Error 1 make[5]: Leaving directory `/disks/chan/mpich2_work/build_tmp/src/mpid/common/datatype' make[4]: *** [all-redirect] Error 2 make[4]: Leaving directory `/disks/chan/mpich2_work/build_tmp/src/mpid/common/datatype' make[3]: *** [all-redirect] Error 2 make[3]: Leaving directory `/disks/chan/mpich2_work/build_tmp/src/mpid/common' make[2]: *** [all-redirect] Error 2 make[2]: Leaving directory `/disks/chan/mpich2_work/build_tmp/src/mpid' make[1]: *** [all-redirect] Error 2 make[1]: Leaving directory `/disks/chan/mpich2_work/build_tmp/src' make: *** [all-redirect] Error 2 ----- "Rajeev Thakur" <thakur(a)mcs.anl.gov> wrote: > Pavan is already en route to Austin. We need to wait at least one > night for > a clean set of tests. So sometime over the weekend or Monday. > Technically we > have until Thursday to release it (our BoF is on Thursday). > > Rajeev > > > > _____ > > From: mpich2-core-bounces(a)mcs.anl.gov > [mailto:[email protected]] On Behalf Of Jayesh Krishna > Sent: Friday, November 14, 2008 12:39 PM > To: balaji(a)mcs.anl.gov > Cc: mpich2-core(a)mcs.anl.gov > Subject: [mpich2-core] Release > > > > Pavan, > When are we cutting the tarball for the release ? > > (PS: Is Hydra good to go ? Did see some commits from you related to > Hydra > last night...) > > Regards, > Jayesh

2 1

Release
by Jayesh Krishna 14 Nov '08

14 Nov '08

Pavan, When are we cutting the tarball for the release ? (PS: Is Hydra good to go ? Did see some commits from you related to Hydra last night...) Regards, Jayesh

3 2